Multiple Kernel Support Vector Regression with Higher Norm in Option Pricing

The purpose of present study is to investigate a nonparametric model that improves accuracy of option prices found by previous models. In this study option prices are calculated using multiple kernel Support Vector Regression with different norm values and their results are compared. L1norm multiple kernel learning Support Vector Regression (MKLSVR) has been successfully applied to option prices forecasting. The advantages of L1-norm multiple kernel learning is that it allows setting of kernel hyperparameter automatically which is one of the most tedious task. L1-norm MKLSVR normally outperforms under all market conditions than single kernel Support Vector Regression. To further minimize forecasting error we adopt Lp-norm multiple kernel Support Vector Regression (p> 1) as it generalize well under all market conditions. In Lp-norm multiple kernel Support Vector Regression (MKLSVR), optimization part is solved using the Sequential Minimal Optimization algorithm (SMO). The Lp-norm SMOMKL SVR is evaluated on forecasting the option prices of European-style Nifty index options in India. Experimental results show that Lp-norm SMOMKL SVR performs better than L1-norm MKL SVR with different methods.